Confidence intervals, standard error, and generalized error rates are typically visualized with "error bars" - thin strokes that are superimposed over the mean. This work presents the results of crowdsourced experiments which illustrate that viewers misinterpret these encodings even at the most basic level (where one would hope larger margins of error reduce the confidence in judgments about means). We then present evaluations of three alternate (or supplemental) visual encodings for the same task and show that choice of visual encoding can result in viewers who make decisions which are better informed by the margins of error.

@InProceedings{CG13,
  author       = "Correll, Michael and Gleicher, Michael",
  title        = "Error Bars Considered Harmful",
  booktitle    = "IEEE Visualization Poster Proceedings",
  month        = "oct",
  year         = "2013",
  publisher    = "IEEE",
  note         = "to appear",
  url          = "http://graphics.cs.wisc.edu/Papers/2013/CG13"
}